Cursor Displacement and Velocity Profiles for Targets in Various Locations
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Cursor Displacement and Velocity Profiles for Targets in Various Locations
H. Keuning-Van Oirschot and A.J.M. Houtsma
IPO, Center for User-System Interaction
P.O. Box 513
5600 MB Eindhoven
The Netherlands
tel: +31 40 2475234
h.keuning@tue.nl
Abstract
In order to create an algorithm, which predicts the endpoint of a cursor movement, an experiment was performed to
reveal characteristics of cursor movements. In the experiment subjects had to click on a button in the center of the
screen after which they moved the cursor onto a target which subsequently appeared. After the target was reached it
disappeared. Targets were radially located at angles of n * 45 degrees (0 ≤ n ≤ 7), and at distances of 150, 250 or 350
pixels (34 pixels/cm). Each of nine subjects executed this experiment three times, using either a mechanical mouse, an
optical trackball or a mechanical force feedback trackball. The variables included in the analyses were curvature,
variability around the mean path and the distance whereon peak velocity occurred. The results show that there is no
need to compensate for a curvature other than zero in the algorithm. As the devices differ in their amount of variability
and the location of the peak velocity one algorithm for each device is preferable. Within one device it could be
considered to have a less accurate prediction or to make a special-case algorithm for direction and distance. To
increase the accuracy even more one could choose to make an algorithm for each person separately.
feedback can be very helpful in this area as well.
1. Introduction Keyson [4] showed that target acquisition with force
feedback was 10-15% faster than without.
Unfortunately, force feedback in this area clearly needs
Technical development and cognitive ergonomical restriction, because getting force feedback every time
research on human-computer interaction have resulted the cursor crosses an object does not make any sense,
in more userfriendly interfaces. More and qualitatively in fact this will be very disturbing. We think that the
better colors and sounds are being used in a more advantages of force feedback in this area are best used
efficient and attractive way. when it helps the user toward an object he/she wants to
Another phenomenon starting to integrate in reach while not being disturbed by other objects.
human-computer interaction is the use of the sense of Therefore the computer needs to know what the users’
touch. This can be done by tactual or force feedback target is in advance.
devices. For instance, playing games with a force For this purpose we want to create an algorithm that
feedback joystick is already very common. Other areas can predict the users' intended target using the initial
wherein tactual feedback is developing rapidly are part of a cursor movement. Tactual feedback can then
medical training (e.g. the PHANToM) and blind be released to help the user moving toward the target
computer users (e.g. the Virtual Reality mouse). and away from non-targets. In order to create such an
In every-day computer use the use of tactual algorithm we need knowledge about the characteristics
feedback is not yet very common. However, tactual of cursor movements toward a target. Characteristicscan be the curvature of paths, the variability of paths − We calculated at what fraction of the target
around the mean path and the velocity profiles. When distance velocity was maximum. In the general
we know these path's statistics it is possible to use the algorithm this fraction sets the multiplication
initially observed direction and velocity of a cursor factor to assess the target distance. As well as the
movement to predict the location of the users' intended other variables, this fraction could be influenced
target. The amount of variability can be used to by direction, distance, devices or user and
estimate the maximum target resolution at which an therefore needs special-case multiplication
algorithm can predict the user's target with a certain factors.
pre-specified probability. However, these
characteristics could be influenced by factors like 2.1. Material and techniques
direction, distance, device and user.
In this paper we describe an experiment in which The experiment was done with three devices; a
we explore cursor movements from a fixed starting mechanical mouse, an optical trackball and a
point to targets located at three different distances and mechanical force feedback trackball. The mechanical
in eight different directions. When the factors mouse we used was a Logitec mouse. We call this a
mentioned do not influence the characteristics it is mechanical mouse because the movement of the cursor
possible to create a general prediction-algorithm which is related to the mechanical movement of the ball
is valid for all cases. When the factors do influence the within the mouse. The optical trackball was the
characteristics, it might be useful to create special-case Logitec Marble Mouse. The force feedback trackball is
algorithms or to give in to the accuracy of the a trackball with two motors, which can roll the ball in
prediction. either x- or y-direction [2]. Without activating the
motors this trackball acts as a normal trackball except
that it is rather big (diameter 55mm) and that the
2. The experiment friction is greater as a result of the contact wheels of
the motors.
This experiment aimed to analyse cursor trajectories
created by computer users in eight different directions,
three different distances and three different devices.
The data sampled were the co-ordinates and time.
From these data we calculated mean paths curvatures,
the variability around the mean paths and velocity
profiles.
In the analyses the following aspects were involved:
− Per device we tested whether the mean path
curvatures in each direction and on each distance
differed from zero. A non-zero curvature would
mean that the trajectories are not straight, but
curved in a certain way. When this is the case it
could be useful to compensate for it in the general
prediction-algorithm. Also, when this is the case
for some of the mean paths, for example in one
direction, it could be useful to alter the general Figure 1. The mechanical force feedback
algorithm for this occasion. trackball (opened)
− We investigated the influence of the different
factors on the variability around the mean path. We used a PC (Pentium II) and a 17-inch Dell-monitor
The amount of variability defines the accuracy with 1248*768 pixels for the sessions with the
with which the prediction can be made. More mechanical mouse and the optical trackball. For the
variability means a less accurate prediction. For session with the force feedback trackball we used a
example when 90% of the paths vary within 20 386-PC and a 17-inch Philips Monitor. On the screen
pixels around the mean path, an accuracy of 90% an interface was created with Visual Basic 5.0. The
can be reached when objects are located 40 pixels target consisted of a black circle 30 pixels in diameter,
away from each other. That is two times the radially located at angles of n * 45 degrees (0 ≤ n ≤ 7),
variation around each mean path. By decreasing and at distances of 150, 250 or 350 pixels (34
the resolution of the objects the accuracy of the pixels/cm).
prediction will increase.During each trial x- and y-co-ordinates (in pixels) and The variability of the paths is represented by the
the system time were sampled. This was done at 50 Hz. standard deviation of all sample points around the
mean path of the ten trajectories performed per subject
2.2. Subjects per direction and per device. The standard deviation is
calculated over each ten-percent section of the whole
Subjects were nine colleagues, five men and four path.
women. The mean age was 29.6 with a standard
deviation of 4.5. All subjects were experienced mouse
users, but had no experience with trackballs. 3. Results
2.3. Procedure Figure 2 shows the mean paths and the variability
around it for each device and each of the three different
In the experiment subjects had to click on a button in distances. Standard deviation bars were computed and
the center of the screen. A target appeared in one of plotted at every one-tenth path length. From this figure
eight directions and at one of three distances to which a number of statistics can be derived for both curvature
the subjects moved the cursor. After the target was and variability.
reached it disappeared. The next trial started when the
button was clicked. 3.1. Curvature
2.4. Design
The mean paths shown in figure 2 look rather straight,
but this can not be simply assumed to be the case (as
The experiment was divided into three sessions. In
Murata [5] did). We tested the calculated curvature
each session another device was used, the order of
against zero for every direction separately. It was also
which was random.
tested separately per device in order to observe
Each session consisted of a practice and an
possible systematic effects of the mechanics of each
experimental part. The practice was included to get the
device.
subject accustomed to the device and the task. In the
Curvature is significantly different from zero on
practice part each target was presented two times in a
almost every direction for almost every device. Paths
random order. In the experimental part each target was
made with the mechanical force feedback trackball are
presented ten times, also randomly. A session lasted
three to four times more curved than paths with the
for about 15 minutes.
other two devices. For the mechanical mouse and the
optical trackball curvatures range from –2 till + 6. The
2.5. Variables optical trackball shows curvature from –8 to +11.
However, including a curvature variable into the
Before calculating the curvature and the variability we prediction algorithm will only be effective when the
performed a resampling on the data, because the curvature is more than the amount of variability.
original sampling was time controlled. This means that Looking at Figure 2, this will not be the case.
velocity of the movement influences the spacing
between the samples. To calculate the curvature of a
3.2. Variability
trajectory, however, we want equal spatial distances
between the samples, because then every part of the
trajectory contributes equally to the calculation of the An overall mean variability of 12.63 pixels was
curvature. Therefore, for each trajectory new samples calculated. A repeated measures ANOVA on
were calculated, by interpolation, for every one-percent variability, with device, distance and direction as
of the total travelled distance (distance of the trajectory within-subjects factors, showed that there is a
itself). significant influence from device (F2,7=25.526, peach device are significantly less variable in horizontal (3) control device: the overall variability is highest
direction than movements in the oblique directions. with the optical trackball (16.3 pixels) and lowest with
Vertical movements are somewhat more variable than the mechanical mouse (9.7 pixels). These results are
horizontal movements, but this effect is not significant. comparable with results found in an earlier experiment
(2) distance: the nearest target results in the least by Keuning and Houtsma [3].
variability (10,6 pixels), except for the mechanical
force feedback trackball which yields a more complex
pattern.
a b c
-400 -400 -400
-200 -200 -200
y, (in pixels)
y, (in pixels)
y, (in pixels)
0 0 0
A
200 200 200
400 400 400
-400 -200 0 200 400 -400 -200 0 200 400 -400 -200 0 200 400
x (in pixels) x (in pixels) x (in pixels)
-400 -400 -400
-200 -200 -200
y, (in pixels)
y, (in p ixe ls)
y, (in pixels)
B 0 0 0
200 200 200
400 400 400
-400 -200 0 200 400 -400 -200 0 200 400 -400 -200 0 200 400
x (in pixels) x (in pixels) x (in pixels)
-400 -400 -400
-200 -200 -200
y, (in pixels)
y, (in pixels)
y, (in pixels)
0 0 0
C
200 200 200
400 400
-400 -200 0 200 400 -400 -200 0 200 400 400
x (in pixels) -400 -200 0 200 400
x (in pixels) x (in pixels)
Figure 2: Mean paths and variability for A. mechanical mouse, B. optical trackball, C. mechanical
force feedback trackball, separately for each distance a. 150 pixels, b. 250 pixels, c. 350 pixels.3.4. Target resolution
3.3. Velocity peak
Assuming variability for the nearest targets of 10.6
Averaged over all directions, distances, control devices pixels at 50% of the distance, a prediction accuracy of
and subjects the peak velocity was reached at 53% of 69% is reached within 7,5 degrees. The farthest targets,
the total distance. A repeated measures ANOVA on having a mean variability of 14.6 pixels, result in a
the location of peak velocity showed that there is a 69% accuracy within 4,5 degrees. Taking the shortest
significant influence of device (F2,7=5.658, pYou can also read
